ABSTRACT
This chapter discusses the general pattern of machine learning and some of the issues that arise. Several specific machine learning methods will be described: deductive learning, inductive learning, and explanation-based learning. A machine learning program can take descriptions of situations couched in terms of these factors and then infer rules that match the expert's behaviour. The chapter looks at several of these issues, which will be important when looking at particular learning algorithms. The new knowledge is thus effectively in the form of several rules. In machine learning, this job of classification is often called concept learning. The equivalent problem where it occurs in discrete systems leads to a sequence of local minima occurring along the line of the ridge. Choosing continuous parameters is beyond the scope of tree-searching techniques. On the hand, traditional mathematical optimization techniques deal only with continuous variables. However, hill climbing can be used with these rather difficult hybrid problems.
